Intel design SYSTEMS no just cpus and chipsets. Therefore instead of cpu manuf. itll be chipset but probably more importantly - video card mauf. Low end will have agp for a long while yet, but ati/nvidia will phase it out within a year, ati moreso. If people are gonna start getting pedantic about what's a chipset, this will get boring fast.
LOL - Wait - why is your inane point less boring than someone else's inane point? And in the end, everything Intel does in its chipset division and its product division is ONLY to drive CPU sales – that’s it – that’s what they consider themselves. Not a chipset manufacturer not a system manufacturer (ever see Intel commercials for either of these two products?) Anyway its all just symantics. They may start making breakfast cereals. LMAO - hold still, imma get you. Actually I think it is more correct to say that I conceded the point that a person could argue that the northbridge is not incorporated into the CPU and if someone has the knowledge and wherewithal to back up their argument then I would let it go. However, for my day to day business and for the day to day busines of people who deal with the northbridge and CPU's daily (except for Via chipsets/engineers) we all agree (well most) that the northbridge is the memory controller and the rest of the world be damned. As far as we give a hoot, AGP is just a fancy schamcy PCI slot.
I don't agree, there's more to the northbridge than the memory controller. It's the chipset that provides the direct interface between the CPU and GPU, via the AGP bus/PCI-Express bus.
let's get one thing clear too, just to avoid any further confusion - PCI-express is not PCI-X PCI-Express is a totally new tech. PCI-X is a new version of the PCI spec, is bacwards compatible to PCI, and will replace PCI.
OK - you can disagree all you want, but I am not going to argue with the engineers at AMD esp when I agree with them. There really are two proper ways to think about this and they both say the NB is gone (OK except for Via which is clinging to the traditional design like a horse carriage). 1-Everything you would have attached to a NB traditionally (Memory, PCI-E, AGP, and the SB), each one of these things can now be connected directly to the CPU (even the SB). If, everything that used to be done in the NB is now done in the CPU then it follows logically to say the NB is now in the CPU. 2-The NB got its name from its position in a system diagram as the component between the CPU and the SB (CPU at the top or left side and the NB above the SB). With the K8, the SB can be directly attached to the CPU so in the diagram there just is no logical place to put something and call it a NB. So Via is living in the past and trying desperately to make themselves feel important, but they are just duplicating effort and there is no need for that. Except for the HW interface (the actual slot / volatages / data paths / signals) PCI, AGP, PCI-X, and PCI-E are the same creature. There is not any more difference between PCI and AGP then there is between PCI and PCI-E. AGP, PCI-X, and PCI-E are all just different versions of PCI with different objectives in mind and thus a different HW interface.
No, VIA aren't living in the past... if they are in the past, explain why they are the performance leaders with their current Athlon 64 chipsets? (not including their K8T890 chipset that was just released) A single chip provides no ability to update the chipset in small parts... If you want to update an NForce 3 chipset, you've got to completely redesign the chipset, which is what they're having to do with CK8-04. This is why VIA have a clearer path in to the PCI-Express market. At the launch of K8T890, there wasn't an updated southbridge on the test system that I saw, but it is to be featured on new and upcoming motherboards. The reason why they do this is explained in the K8T890 article that Bindibadgi wrote - functionality is the key here. There's a hell of a lot more testing required in the southbridge than in the northbridge. Combining both means that there's double the amount of testing to do every time you want to refresh the chipset. VIA's way means that you only have to test either northbridge or southbridge separately, depending on which part of the chipset you are wanting to refresh. It also means backward compatability - are we going to see any NForce-based PCI-Express chipsets? I highly doubt it. The single chip does offer one benefit, and that is that there is no latency between the north and south bridges, as they're the same chip in the case of NForce 3 and (I would imagine having not seen the chipset yet...) NForce 4. I can't see how you come to the conclusion that AGP and PCI are the same as the PCI-Express interface. How can a unidirectional interface be "basically the same as" a bidirectional interface? I'm intregued as to how you see them being "basically the same". If the hardware interface is different, doesn't that essentially mean that they're a different thing? Sure, they're a different implementation of the same task, but they're not basically the same by any stretch of the imagination IMHO.
Well, its like this . . . Regardless of any benchmarks, the fact is anything that was once attached to the NorthBridge can now be attached directly to the K8 CPU. Thus, all functionality of the NorthBridge can be done in the K8 CPU. Thus, a NorthBridge is built into the K8 CPU. So, if someone wants to attach other components and call it a NorthBridge fine. But this does not preclude the fact that there is already a NorthBridge built into the K8. As for the PCI, AGP, PCI-X, PCI-E thing . . . well they are obviously not exactly the same. I am saying that each one is simply a compatible version of PCI. Only the hardware inter-links are different from one to another. The control methods are identical but updated. Thus, if I designed a PCI card in 1991 when PCI first got its go, and I wrote a driver for that card; I could re-use the exact same driver in an AGP version, a PCI-X version, and a PCI-E version. I could do it without changing a single line of code. Granted, it wouldn't take advantage of all the new options available, but the only thing I would HAVE to do is update the interconnecting hardware for the new slot. So, if you are a hardware guy, they seem vastly different, but to a software guy moving from PCI to AGP is the same thing as moving from AX to EAX (sorry, if you don’t know assembly this is the best example I can think of) or from PCI to PCI-E. Thus my assertion. Sorry if I sound like a jerk. I mean I am a jerk, but I don't mean to come accross as one. EDIT: HA - this needs to be my SIG
It seems to me, that the NB isn't built onto the CPU, the chipset just allows it so that there is no need for a physical NB. There is no northbridge on the CPU, just jobs of the northbridge have been passed onto the CPU. As for PCI, AGP, etc. I think you have been proved wrong and come out claiming some obscure link between them... It's like claiming any bus/interface is the same, apart from pretty much everything?
Haha - OK - well, I am saying goto the AMD website then think again. If that don't work, then talk to an AMD engineer. If you don't buy my explanation of the compatibility of the different busses (and AGP) - try reading it again. It’s about as clear as I can make it. OK - maybe try this thought though it may confuse the issue . . . TCP/IP works because it has different layers and different protocols for each layer. Thus the top layer which is used say by IE doesn't care if it is traveling over the phone line or an Ethernet connection. However, in PCI there are only 3 layers. Thus, when you go from PCI-E to PCI to AGP to PCI-X, the only thing that changes is the HW layer. Thus, software designed to run on your PCI card can run on your AGP card or your PCI-E card, etc. The same way that you can run IE over your phone or your Ehternet connection. Because all three layers come from the same specification (the PCI spec), changing from one format to another (AGP, PCI-E) is just a matter of choosing one implementation out of the bunch. But in the end, the entire thing is controled by one specification (The PCI spec) which subordinates the other specs. Man - i think I made it worse not better.
Ah - ok, so you don't think it is significant that you can use Internet Explorer on your phone line or your wireless or your DSL or your Ethernet. These are just hardware layers to the TCP/IP spec. PCI-E, PCI-X, AGP are just hardware layers to the PCI spec.
TCP/IP has got SFA to do with this. What is "PCI" spec then? As BigZ says, some of the standards handle data in one way, others can handle data bidirectionally. I still don't see how they are "exactly the same", oh, except for everything about them.
sigh From the PCI-SIG website. The PCI-SIG wrote and maintain PCI, PCI-X, and PCI-E. If a device is PCI, PCI-X, or PCI-E compliant, that means that it conforms to the specifications set forth by the PCI-SIG. No one has authority over them on these matters. More from the PCI-SIG website: From the PCI-E FAQ section: Please take note that the very first feature they list is that it is compatable with PCI. From the PCI-X FAQ section: Please take note that again the very first feature they list is its compatability with PCI. As for AGP, it is still "owned" by Intel and they have pulled almost everything regarding AGP (in the face of PCI-E). But with a little work, you can find the AGP 2.0 specification, in which you will notice that AGP is based entirely upon PCI (as is PCI-X and PCI-E). So I challenge you to read the AGP specification. It will go into detail about the 4 extensions the AGP specification makes to the PCI specification. That’s right, you count them right - the entire AGP specification is considered merely a 4 point extension to the PCI specification. Do they seem any more related to one another now? I mean I tried to be less vague this time. So, just because the little slot thingy looks different for each one, the people that designed them went out of their way to make sure that internaly they were 100% backwards compatable to PCI. They even brag about it. Now as to the idea that one of these technologies may not be bi-directional - umm no. Not possible. How is it they say? Pwned
interesting. but i seem to recall reading somewhere that PCI-express wasn't PCI backwards compatible. i even posted as such(which i now notice you replied to). ah well, you win some, you lose some, you learn some.
The Hardware is not (the slot), but the Software is (the drivers). Which is actually a mean feat. You can't take a driver from SCSI to IDE or from any of them to USB. But, a driver writen for PCI will work on any of these systems (PCI-X, PCI-E, AGP).
It's an onboard memory controller, and that's what I've always heard it referred to as. The "Northbridge" on the K8 does not support the PCI-Express bus (I know, you haven't said this... but hold tight... I'll get to my point, eventually). Also while we're on this thought path, the traditional north bridge has always (well, in recent times, as they say) supported the AGP interface. Yes I know that the i875P chipset (as an example), like most chipsets, used the southbridge to support the PCI bus with a single 133MB/s channel... with the evolution of the PCI-Express interface in conjunction with the onboard memory controller that is located on the K8, everything can be handled by the southbridge so to speak. With the AGP 8x interface, there was/is still a requirement for a "northbridge" as you refer to it. The K8 does not handle data from the AGP bus directly - it's taken through the chipset. Traditionally, the Northbridge has handled the AGP channel. What I am trying to prove here, is that the Northbridge hasn't completely moved on to the K8's die. It's only the memory controller that has moved that way; this is where the evolution of the HyperTransport bus came from - it's the CPU's direct channel to local memory. VIA's K8 chipset implementations are perfectly valid, they allow for an element of mix and match with different northbridge and southbridges... maybe some motherboard manufacturers would like to use the similar performing K8T800 Pro chipset with the new VT8251 southbridge, instead of the K8T890, to allow for two PCIe x1 slots, but no PEG slot to replace the AGP 8x slot? That is the beauty of the way that VIA manufactures their chipsets. I understand that now, and I do have reasonable knowledge of assembler - unfortunately, it's been 3 or so years since I last touched it. Thanks for explaining it from another POV than my normalised HW POV.
LOL - well ok, I have had my fill on this go around. I know I am hard on VIA sometimes, but honestly I try real hard not to. Seems like most of my stuff is VIA these days so maybe I should leave them alone.
hell yeah, that's some badass design going on there. i should spend some time reading about this stuff - i've gone a bit out of touch (we studied the MIPS datapath last year, but nothing too in-depth)
